The Token Games: Evaluating Language Model Reasoning with Puzzle Duels

In three linesTTG (Token Games) is an evaluation framework where language models challenge each other by creating programming puzzles. The system uses pairwise duels and Elo ratings to compare 10 frontier models. Results match existing benchmarks (Humanity's Last Exam) for under $200 USD without human puzzle curation.

Summary generated by Claude — human-verified